Earwax
Your example is exactly the point. Jimi is playing his gutar, effected the way he wants, and recording same, all in real time.
Hm.
"Once it was in the can, then you'd start mucking with it.... For the various effects and things..." (Chas Chandler on recording Jimi Hendrix).
Recording Jimi Hendrix was not a simple process either in the studio or live. Even live recordings would be post-processed in the studio prior to release if for no other reason than to sort out the mix, get things to the right length to fit on a side on a LP and to master for the requirements of vinyl.
As for recording soft synth's audio output rather than recording the MIDI then bouncing/freezing the synth I really don't see the issue unless you are dealing with a synth that has functions that can't be MIDI controlled.
If a synth (or e.g. Sonar's step sequencer) has a randomising factor, then whether that first take is going to be the best is a matter of luck in any case. My approach is to do multiple bounces of the part to audio then pick between them and even comp the best bits together.
Other than that, when you play a software (or hardware) synth using a MIDI controller the synth is responding to the MIDI. The synth has no way to tell if the MIDI data is being generated as part of a "live" performance or sent after the fact from a sequencer or Sonar MIDI track..